rumale 0.8.0

data/lib/rumale/evaluation_measure/recall.rb

@@ -0,0 +1,50 @@
+# frozen_string_literal: true
+
+require 'rumale/base/evaluator'
+require 'rumale/evaluation_measure/precision_recall'
+
+module Rumale
+  # This module consists of the classes for model evaluation.
+  module EvaluationMeasure
+    # Recall is a class that calculates the recall of the predicted labels.
+    #
+    # @example
+    #   evaluator = Rumale::EvaluationMeasure::Recall.new
+    #   puts evaluator.score(ground_truth, predicted)
+    class Recall
+      include Base::Evaluator
+      include EvaluationMeasure::PrecisionRecall
+
+      # Return the average type for calculation of recall.
+      # @return [String] ('binary', 'micro', 'macro')
+      attr_reader :average
+
+      # Create a new evaluation measure calculater for recall score.
+      #
+      # @param average [String] The average type ('binary', 'micro', 'macro')
+      def initialize(average: 'binary')
+        check_params_string(average: average)
+        @average = average
+      end
+
+      # Calculate average recall
+      #
+      # @param y_true [Numo::Int32] (shape: [n_samples]) Ground truth labels.
+      # @param y_pred [Numo::Int32] (shape: [n_samples]) Predicted labels.
+      # @return [Float] Average recall
+      def score(y_true, y_pred)
+        check_label_array(y_true)
+        check_label_array(y_pred)
+
+        case @average
+        when 'binary'
+          recall_each_class(y_true, y_pred).last
+        when 'micro'
+          micro_average_recall(y_true, y_pred)
+        when 'macro'
+          macro_average_recall(y_true, y_pred)
+        end
+      end
+    end
+  end
+end

data/lib/rumale/kernel_approximation/rbf.rb

@@ -0,0 +1,121 @@
+# frozen_string_literal: true
+
+require 'rumale/utils'
+require 'rumale/base/base_estimator'
+require 'rumale/base/transformer'
+
+module Rumale
+  # Module for kernel approximation algorithms.
+  module KernelApproximation
+    # Class for RBF kernel feature mapping.
+    #
+    # @example
+    #   transformer = Rumale::KernelApproximation::RBF.new(gamma: 1.0, n_coponents: 128, random_seed: 1)
+    #   new_training_samples = transformer.fit_transform(training_samples)
+    #   new_testing_samples = transformer.transform(testing_samples)
+    #
+    # *Refernce*:
+    # 1. A. Rahimi and B. Recht, "Random Features for Large-Scale Kernel Machines," Proc. NIPS'07, pp.1177--1184, 2007.
+    class RBF
+      include Base::BaseEstimator
+      include Base::Transformer
+
+      # Return the random matrix for transformation.
+      # @return [Numo::DFloat] (shape: [n_features, n_components])
+      attr_reader :random_mat
+
+      # Return the random vector for transformation.
+      # @return [Numo::DFloat] (shape: [n_components])
+      attr_reader :random_vec
+
+      # Return the random generator for transformation.
+      # @return [Random]
+      attr_reader :rng
+
+      # Create a new transformer for mapping to RBF kernel feature space.
+      #
+      # @param gamma [Float] The parameter of RBF kernel: exp(-gamma * x^2).
+      # @param n_components [Integer] The number of dimensions of the RBF kernel feature space.
+      # @param random_seed [Integer] The seed value using to initialize the random generator.
+      def initialize(gamma: 1.0, n_components: 128, random_seed: nil)
+        check_params_float(gamma: gamma)
+        check_params_integer(n_components: n_components)
+        check_params_type_or_nil(Integer, random_seed: random_seed)
+        check_params_positive(gamma: gamma, n_components: n_components)
+        @params = {}
+        @params[:gamma] = gamma
+        @params[:n_components] = n_components
+        @params[:random_seed] = random_seed
+        @params[:random_seed] ||= srand
+        @random_mat = nil
+        @random_vec = nil
+        @rng = Random.new(@params[:random_seed])
+      end
+
+      # Fit the model with given training data.
+      #
+      # @overload fit(x) -> RBF
+      #
+      # @param x [Numo::NArray] (shape: [n_samples, n_features]) The training data to be used for fitting the model.
+      #   This method uses only the number of features of the data.
+      # @return [RBF] The learned transformer itself.
+      def fit(x, _y = nil)
+        check_sample_array(x)
+
+        n_features = x.shape[1]
+        @params[:n_components] = 2 * n_features if @params[:n_components] <= 0
+        @random_mat = Rumale::Utils.rand_normal([n_features, @params[:n_components]], @rng) * (2.0 * @params[:gamma])**0.5
+        n_half_components = @params[:n_components] / 2
+        @random_vec = Numo::DFloat.zeros(@params[:n_components] - n_half_components).concatenate(
+          Numo::DFloat.ones(n_half_components) * (0.5 * Math::PI)
+        )
+        self
+      end
+
+      # Fit the model with training data, and then transform them with the learned model.
+      #
+      # @overload fit_transform(x) -> Numo::DFloat
+      #
+      # @param x [Numo::DFloat] (shape: [n_samples, n_features]) The training data to be used for fitting the model.
+      # @return [Numo::DFloat] (shape: [n_samples, n_components]) The transformed data
+      def fit_transform(x, _y = nil)
+        check_sample_array(x)
+
+        fit(x).transform(x)
+      end
+
+      # Transform the given data with the learned model.
+      #
+      # @overload transform(x) -> Numo::DFloat
+      #
+      # @param x [Numo::DFloat] (shape: [n_samples, n_features]) The data to be transformed with the learned model.
+      # @return [Numo::DFloat] (shape: [n_samples, n_components]) The transformed data.
+      def transform(x)
+        check_sample_array(x)
+
+        n_samples, = x.shape
+        projection = x.dot(@random_mat) + @random_vec.tile(n_samples, 1)
+        Numo::NMath.sin(projection) * ((2.0 / @params[:n_components])**0.5)
+      end
+
+      # Dump marshal data.
+      # @return [Hash] The marshal data about RBF.
+      def marshal_dump
+        { params: @params,
+          random_mat: @random_mat,
+          random_vec: @random_vec,
+          rng: @rng }
+      end
+
+      # Load marshal data.
+      # @return [nil]
+      def marshal_load(obj)
+        @params = obj[:params]
+        @random_mat = obj[:random_mat]
+        @random_vec = obj[:random_vec]
+        @rng = obj[:rng]
+        nil
+      end
+    end
+  end
+end

data/lib/rumale/kernel_machine/kernel_svc.rb

@@ -0,0 +1,193 @@
+# frozen_string_literal: true
+
+require 'rumale/base/base_estimator'
+require 'rumale/base/classifier'
+require 'rumale/probabilistic_output'
+
+module Rumale
+  # This module consists of the classes that implement kernel method-based estimator.
+  module KernelMachine
+    # KernelSVC is a class that implements (Nonlinear) Kernel Support Vector Classifier
+    # with stochastic gradient descent (SGD) optimization.
+    # For multiclass classification problem, it uses one-vs-the-rest strategy.
+    #
+    # @example
+    #   training_kernel_matrix = Rumale::PairwiseMetric::rbf_kernel(training_samples)
+    #   estimator =
+    #     Rumale::KernelMachine::KernelSVC.new(reg_param: 1.0, max_iter: 1000, random_seed: 1)
+    #   estimator.fit(training_kernel_matrix, traininig_labels)
+    #   testing_kernel_matrix = Rumale::PairwiseMetric::rbf_kernel(testing_samples, training_samples)
+    #   results = estimator.predict(testing_kernel_matrix)
+    #
+    # *Reference*
+    # 1. S. Shalev-Shwartz, Y. Singer, N. Srebro, and A. Cotter, "Pegasos: Primal Estimated sub-GrAdient SOlver for SVM," Mathematical Programming, vol. 127 (1), pp. 3--30, 2011.
+    class KernelSVC
+      include Base::BaseEstimator
+      include Base::Classifier
+
+      # Return the weight vector for Kernel SVC.
+      # @return [Numo::DFloat] (shape: [n_classes, n_trainig_sample])
+      attr_reader :weight_vec
+
+      # Return the class labels.
+      # @return [Numo::Int32] (shape: [n_classes])
+      attr_reader :classes
+
+      # Return the random generator for performing random sampling.
+      # @return [Random]
+      attr_reader :rng
+
+      # Create a new classifier with Kernel Support Vector Machine by the SGD optimization.
+      #
+      # @param reg_param [Float] The regularization parameter.
+      # @param max_iter [Integer] The maximum number of iterations.
+      # @param probability [Boolean] The flag indicating whether to perform probability estimation.
+      # @param random_seed [Integer] The seed value using to initialize the random generator.
+      def initialize(reg_param: 1.0, max_iter: 1000, probability: false, random_seed: nil)
+        check_params_float(reg_param: reg_param)
+        check_params_integer(max_iter: max_iter)
+        check_params_boolean(probability: probability)
+        check_params_type_or_nil(Integer, random_seed: random_seed)
+        check_params_positive(reg_param: reg_param, max_iter: max_iter)
+        @params = {}
+        @params[:reg_param] = reg_param
+        @params[:max_iter] = max_iter
+        @params[:probability] = probability
+        @params[:random_seed] = random_seed
+        @params[:random_seed] ||= srand
+        @weight_vec = nil
+        @prob_param = nil
+        @classes = nil
+        @rng = Random.new(@params[:random_seed])
+      end
+
+      # Fit the model with given training data.
+      #
+      # @param x [Numo::DFloat] (shape: [n_training_samples, n_training_samples])
+      #   The kernel matrix of the training data to be used for fitting the model.
+      # @param y [Numo::Int32] (shape: [n_training_samples]) The labels to be used for fitting the model.
+      # @return [KernelSVC] The learned classifier itself.
+      def fit(x, y)
+        check_sample_array(x)
+        check_label_array(y)
+        check_sample_label_size(x, y)
+
+        @classes = Numo::Int32[*y.to_a.uniq.sort]
+        n_classes = @classes.size
+        _n_samples, n_features = x.shape
+
+        if n_classes > 2
+          @weight_vec = Numo::DFloat.zeros(n_classes, n_features)
+          @prob_param = Numo::DFloat.zeros(n_classes, 2)
+          n_classes.times do |n|
+            bin_y = Numo::Int32.cast(y.eq(@classes[n])) * 2 - 1
+            @weight_vec[n, true] = binary_fit(x, bin_y)
+            @prob_param[n, true] = if @params[:probability]
+                                     Rumale::ProbabilisticOutput.fit_sigmoid(x.dot(@weight_vec[n, true].transpose), bin_y)
+                                   else
+                                     Numo::DFloat[1, 0]
+                                   end
+          end
+        else
+          negative_label = y.to_a.uniq.min
+          bin_y = Numo::Int32.cast(y.ne(negative_label)) * 2 - 1
+          @weight_vec = binary_fit(x, bin_y)
+          @prob_param = if @params[:probability]
+                          Rumale::ProbabilisticOutput.fit_sigmoid(x.dot(@weight_vec.transpose), bin_y)
+                        else
+                          Numo::DFloat[1, 0]
+                        end
+        end
+
+        self
+      end
+
+      # Calculate confidence scores for samples.
+      #
+      # @param x [Numo::DFloat] (shape: [n_testing_samples, n_training_samples])
+      #     The kernel matrix between testing samples and training samples to compute the scores.
+      # @return [Numo::DFloat] (shape: [n_testing_samples, n_classes]) Confidence score per sample.
+      def decision_function(x)
+        check_sample_array(x)
+
+        x.dot(@weight_vec.transpose)
+      end
+
+      # Predict class labels for samples.
+      #
+      # @param x [Numo::DFloat] (shape: [n_testing_samples, n_training_samples])
+      #     The kernel matrix between testing samples and training samples to predict the labels.
+      # @return [Numo::Int32] (shape: [n_testing_samples]) Predicted class label per sample.
+      def predict(x)
+        check_sample_array(x)
+
+        return Numo::Int32.cast(decision_function(x).ge(0.0)) * 2 - 1 if @classes.size <= 2
+
+        n_samples, = x.shape
+        decision_values = decision_function(x)
+        Numo::Int32.asarray(Array.new(n_samples) { |n| @classes[decision_values[n, true].max_index] })
+      end
+
+      # Predict probability for samples.
+      #
+      # @param x [Numo::DFloat] (shape: [n_testing_samples, n_training_samples])
+      #     The kernel matrix between testing samples and training samples to predict the labels.
+      # @return [Numo::DFloat] (shape: [n_samples, n_classes]) Predicted probability of each class per sample.
+      def predict_proba(x)
+        check_sample_array(x)
+
+        if @classes.size > 2
+          probs = 1.0 / (Numo::NMath.exp(@prob_param[true, 0] * decision_function(x) + @prob_param[true, 1]) + 1.0)
+          return (probs.transpose / probs.sum(axis: 1)).transpose
+        end
+
+        n_samples, = x.shape
+        probs = Numo::DFloat.zeros(n_samples, 2)
+        probs[true, 1] = 1.0 / (Numo::NMath.exp(@prob_param[0] * decision_function(x) + @prob_param[1]) + 1.0)
+        probs[true, 0] = 1.0 - probs[true, 1]
+        probs
+      end
+
+      # Dump marshal data.
+      # @return [Hash] The marshal data about KernelSVC.
+      def marshal_dump
+        { params: @params,
+          weight_vec: @weight_vec,
+          prob_param: @prob_param,
+          classes: @classes,
+          rng: @rng }
+      end
+
+      # Load marshal data.
+      # @return [nil]
+      def marshal_load(obj)
+        @params = obj[:params]
+        @weight_vec = obj[:weight_vec]
+        @prob_param = obj[:prob_param]
+        @classes = obj[:classes]
+        @rng = obj[:rng]
+        nil
+      end
+
+      private
+
+      def binary_fit(x, bin_y)
+        # Initialize some variables.
+        n_training_samples = x.shape[0]
+        rand_ids = []
+        weight_vec = Numo::DFloat.zeros(n_training_samples)
+        # Start optimization.
+        @params[:max_iter].times do |t|
+          # random sampling
+          rand_ids = [*0...n_training_samples].shuffle(random: @rng) if rand_ids.empty?
+          target_id = rand_ids.shift
+          # update the weight vector
+          func = (weight_vec * bin_y).dot(x[target_id, true].transpose).to_f
+          func *= bin_y[target_id] / (@params[:reg_param] * (t + 1))
+          weight_vec[target_id] += 1.0 if func < 1.0
+        end
+        weight_vec * bin_y
+      end
+    end
+  end
+end

data/lib/rumale/linear_model/base_linear_model.rb

@@ -0,0 +1,89 @@
+# frozen_string_literal: true
+
+require 'rumale/base/base_estimator'
+require 'rumale/optimizer/nadam'
+
+module Rumale
+  module LinearModel
+    # BaseLinearModel is an abstract class for implementation of linear estimator
+    # with mini-batch stochastic gradient descent optimization.
+    # This class is used for internal process.
+    class BaseLinearModel
+      include Base::BaseEstimator
+
+      # Initialize a linear estimator.
+      #
+      # @param reg_param [Float] The regularization parameter.
+      # @param fit_bias [Boolean] The flag indicating whether to fit the bias term.
+      # @param bias_scale [Float] The scale of the bias term.
+      # @param max_iter [Integer] The maximum number of iterations.
+      # @param batch_size [Integer] The size of the mini batches.
+      # @param optimizer [Optimizer] The optimizer to calculate adaptive learning rate.
+      #   If nil is given, Nadam is used.
+      # @param random_seed [Integer] The seed value using to initialize the random generator.
+      def initialize(reg_param: 1.0, fit_bias: false, bias_scale: 1.0,
+                     max_iter: 1000, batch_size: 10, optimizer: nil, random_seed: nil)
+        @params = {}
+        @params[:reg_param] = reg_param
+        @params[:fit_bias] = fit_bias
+        @params[:bias_scale] = bias_scale
+        @params[:max_iter] = max_iter
+        @params[:batch_size] = batch_size
+        @params[:optimizer] = optimizer
+        @params[:optimizer] ||= Optimizer::Nadam.new
+        @params[:random_seed] = random_seed
+        @params[:random_seed] ||= srand
+        @weight_vec = nil
+        @bias_term = nil
+        @rng = Random.new(@params[:random_seed])
+      end
+
+      private
+
+      def partial_fit(x, y)
+        # Expand feature vectors for bias term.
+        samples = @params[:fit_bias] ? expand_feature(x) : x
+        # Initialize some variables.
+        n_samples, n_features = samples.shape
+        rand_ids = [*0...n_samples].shuffle(random: @rng)
+        weight = Numo::DFloat.zeros(n_features)
+        optimizer = @params[:optimizer].dup
+        # Optimization.
+        @params[:max_iter].times do |_t|
+          # Random sampling
+          subset_ids = rand_ids.shift(@params[:batch_size])
+          rand_ids.concat(subset_ids)
+          sub_samples = samples[subset_ids, true]
+          sub_targets = y[subset_ids]
+          # Update weight.
+          loss_gradient = calc_loss_gradient(sub_samples, sub_targets, weight)
+          next if loss_gradient.ne(0.0).count.zero?
+          weight = calc_new_weight(optimizer, sub_samples, weight, loss_gradient)
+        end
+        split_weight(weight)
+      end
+
+      def calc_loss_gradient(_x, _y, _weight)
+        raise NotImplementedError, "#{__method__} has to be implemented in #{self.class}."
+      end
+
+      def calc_new_weight(optimizer, x, weight, loss_gradient)
+        weight_gradient = x.transpose.dot(loss_gradient) / @params[:batch_size] + @params[:reg_param] * weight
+        optimizer.call(weight, weight_gradient)
+      end
+
+      def expand_feature(x)
+        n_samples = x.shape[0]
+        Numo::NArray.hstack([x, Numo::DFloat.ones([n_samples, 1]) * @params[:bias_scale]])
+      end
+
+      def split_weight(weight)
+        if @params[:fit_bias]
+          [weight[0...-1].dup, weight[-1]]
+        else
+          [weight, 0.0]
+        end
+      end
+    end
+  end
+end